基于用户兴趣的可信购物推荐服务系统实现

发布时间：2018-05-29 19:32

本文选题：协同过滤 + 网络购物　；参考：《东南大学》2017年硕士论文

【摘要】：随着互联网的发展,网络商品各色各样,种类越来越繁多,在网络购物多重不安定因素下,人们在网络有效购物也越来越困难,购物效率也会随之下降,制约我国电子商务的发展。随着网上信息数量和商品种类的急速增长对推荐系统提出了严峻的挑战,基于用户的协同过滤推荐中的用户兴趣的定位问题和购买产品的风险评估问题急待解决。本文对传统的基于用户的协同过滤算法改进,形成一套基于用户兴趣的可信购物推荐服务系统。该方法通过对用户兴趣元数据进行分析形成用户兴趣分类,产生不同兴趣类的数据集,通过对不同类的数据集分别进行协同过滤算法训练,产生各个兴趣类对应的最优模型,进而对目标用户未评分的项目进行预测,最后对预测推荐的项目采用信用风险评判并过滤掉高风险项目,实现推荐的项目是可信的。本文的主要工作有以下几个方面:1)设计了基于用户兴趣的可信购物推荐服务系统架构,主要由数据预处理子系统、推荐子系统、交互控制系统、存储子系统、数据后处理子系统、确认子系统组成。2)设计了用户兴趣树,用于从空间和时间维度上描述用户兴趣;实现了兴趣树的构造和更新算法。3)对当前协同过滤算法进行适当改进,设计并实现了推荐子系统,提高了推荐商品的质量和效率;根据推荐系统的输入和输出要素设计并实现了支撑用户人机交互的交互控制子系统。4)设计并实现了基于兴趣树的推荐模型训练方法,可基于用户兴趣树对用户商品聚类,并通过协同过滤训练算法训练模型。5)综合评估用户商品购买风险,设计并实现了风险过滤方法。6)设计了推荐系统测试方案,并对系统进行了验证。测试数据来源为网上正常交易商品的数据;测试结果显示,本系统性能和推荐质量相比较传统推荐系统有所提升,并能过滤其中的风险欺诈信息。
[Abstract]:With the development of the Internet, the variety and variety of online goods are becoming more and more diverse. Under the factors of multiple instability of online shopping, it is becoming more and more difficult for people to shop effectively on the Internet, and the efficiency of shopping will also decline. Restrict the development of electronic commerce of our country. With the rapid growth of the quantity of information and the types of goods on the Internet, the recommendation system is facing a severe challenge. It is urgent to solve the problem of the location of user interest and the risk assessment of purchasing products in the collaborative filtering recommendation based on users. In this paper, the traditional collaborative filtering algorithm based on users is improved to form a trusted shopping recommendation service system based on user interest. In this method, user interest classification is formed by analyzing user interest metadata, and the data sets of different interest classes are generated. By training the different classes of data sets with collaborative filtering algorithm, the optimal model of each interest class is generated. Finally, credit risk evaluation and filtering out high risk items are used to predict the items that are not graded by the target users, and the items that are recommended are credible. The main work of this paper is as follows: 1) the framework of trusted shopping recommendation service system based on user interest is designed, which consists of data preprocessing subsystem, recommendation subsystem, interactive control system, storage subsystem, etc. The user interest tree is designed to describe user interest in space and time dimension, and the interest tree construction and update algorithm .3) is implemented to improve the current collaborative filtering algorithm. The recommendation subsystem is designed and implemented to improve the quality and efficiency of the recommended goods. According to the input and output elements of the recommendation system, the interactive control subsystem. 4) which supports the user's human-computer interaction, is designed and implemented. The recommendation model training method based on the interest tree is designed and implemented. The user commodity can be clustered based on the user's interest tree. The collaborative filtering training algorithm training model. 5) is used to evaluate the risk of commodity purchase. The risk filtering method .6) is used to design and test the recommendation system, and the system is verified. The test results show that the performance and quality of the system are improved compared with the traditional recommendation system and can filter the information of risk fraud.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】